Empirical likelihood for the two-sample mean problem

We apply empirical likelihood method to constructing confidence regions for the difference of the means of two d-dimensional samples. It is shown that the empirical likelihood ratio test has an asymptotic chi-squared distribution. The Bartlett correction for the univariate case (d=1) has been investigated by Jing [1995. Two-sample empirical likelihood method. Statist. Probab. Lett. 24, 315-319]. Unfortunately, the Bartlett correction obtained in that article was incorrect. In this article the correct Bartlett correction is found and its effectiveness is shown by a simulation study.